A Multi-task Platform for Profiling Cognitive and Motivational Constructs in Humans and Nonhuman Primates

Abstract

Background: Understanding the neurobiological substrates of psychiatric disorders requires comprehensive evaluations of cognitive and motivational functions in preclinical research settings. The translational validity of such evaluations will be supported by (1) tasks with high construct validity that are engaging and easy to teach to human and nonhuman participants, (2) software that enables efficient switching between multiple tasks in single sessions, (3) software that supports tasks across a broad range of physical experimental setups, and (4) by platform architectures that are easily extendable and customizable to encourage future optimization and development.

New method: We describe the Multi-task Universal Suite for Experiments (M-USE), a software platform designed to meet these requirements. It leverages the Unity video game engine and C# programming language to (1) support immersive and engaging tasks for humans and nonhuman primates, (2) allow experimenters or participants to switch between multiple tasks within-session, (3) generate builds that function across computers, tablets, and websites, and (4) is freely available online with documentation and tutorials for users and developers. M-USE includes a task library with seven pre-existing tasks assessing cognitive and motivational constructs of perception, attention, working memory, cognitive flexibility, motivational and affective self-control, relational long-term memory, and visuo-spatial problem solving.

Results: M-USE was used to test NHPs on up to six tasks per session, all available as part of the Task Library, and to extract performance metrics for all major cognitive and motivational constructs spanning the Research Domain Criteria (RDoC) of the National Institutes of Mental Health.

Comparison with existing methods: Other experiment design and control systems exist, but do not provide the full range of features available in M-USE, including a pre-existing task library for cross-species assessments; the ability to switch seamlessly between tasks in individual sessions; cross-platform build capabilities; license-free availability; and its leveraging of video-engine capabilities used to gamify tasks.

Conclusions: The new multi-task platform facilitates cross-species translational research for understanding the neurobiological substrates of higher cognitive and motivational functions.

Keywords: Diagnostic Battery; naturalistic behavior; nonhuman primates; video-engine; virtual reality.

Figure 1.. Build architecture and state system of the Multi-task Unified Suite for Experiments (M-USE).

(A) M-USE builds consist of Unity’s native processes (in black), the M_USE State Machine that controls task operations (orange), a Task Library with pre-configured tasks that can be accessed in a plug-and-play manner by the State Machine (red), and custom Modules that govern I/O and other common experimental needs (blue). All interactions with other CPU processes or experimental equipment are mediated through the Modules, allowing the State Machine and Task Library to remain encapsulated. (B) The primary components of the M-USE State Machine include sequences of states at the Session, Task and Trial levels. A developer’s primary work consists in defining the custom Trial states that control their task (in yellow).

Figure 2.

Basic I/O functions of M-USE. (A) Modules control all aspects of I/O, including (B) reading configuration files that apply across a session, or to individual tasks, (C) writing data files for the session and task, (D) receiving participant input, (E) communicating to and from time synchronization devices such as our Arduino-based SyncBox, (F) communicating with eyetrackers, if used, and (G) displaying up to date information and allowing experimenter to manipulate trial variables on an Experimenter Display.

Figure 3.. Temporally precise reconstruction of frame onsets.

To precisely reconstruct events to the physical monitor frame at which events they occurred during task performance, photodiodes can be used to track white-black flashing sequences. Analysis scripts are provided on github to detect the measured onsets (A) and adjust the onset times in case frames were stuck of delayed (B). The time intervals between flashes on the right diode (C) and left (D) diode are precisely adjusted (E).

Figure 4.. Overview of tasks

(A) M-USE state systems incorporates tasks that were chosen to assess functional domains realized by separate subfield of the prefrontal cortex according to Passingham (2021) who distinguished five cortical subfields (left) and primary functional domains. We measure these domains with five primary tasks and two multidimensional tasks (right). (B) Depiction of the multi-task selection screen subjects use to choose tasks (top middle), and illustrations of the tasks and task rules.

Figure 5. Performance Metrics of pre-configured M-USE tasks.

The pre-configured tasks of M-USE (columns) measure a wide array of performance metrics (rows). The metrics map onto six different cognitive affective domains that encompass RDoC constructs (rightmost column). How the metrics are calculated is described in the main text and realized in matlab scripts (The MathWorks, Inc.) accompanying the M-USE github (https://github.com/Multitask-Universal-Suite-for-Expts).

Figure 6. Extracting performance metrics from monkeys performing ≥3 of the pre-configured tasks per daily session.

(A) Visual search regression slopes from 200 trials of a single session showing increased distractor interference (i.e. a conjunction search effect) evident in longer reaction times with more distractors (x-axis) sharing perceptual similarity with the target (red line: r=0.7 regression slope). Distractors that are dissimilar to the target cause a pop-out effect (blue line). (B) Working Memory: Delayed-match-to-sample performance across seven sessions using 150 trials each shows a drop in accuracy at 1.75 s delay and purer performance when the test stimulus is shown with perceptually similar distracting stimuli. (C) Maze Learning: Average proportions of perseverations and overall errors per ‘tile’ over 22 mazes reduce when the monkey repeats the same maze. Mazes pathlength varied from 5 to 12 tiles. The result indexes spatial learning and successful error monitoring. (D) Flexible-Learning: Average learning curves (left) over seven sessions indicate slower learning (more trials to criterion performance) for extra- than intra-dimensional block switches of target features (right). (E) Effort Control: Example results from a block of 80 trials of the Effort Control task (left panel) shows the monkey chose generally more likely the option with higher reward outcome (y-axis values are above 0.5) and that his reward sensitivity increases when the absolute difference of effort between options is larger, indexing he resolves a conflict between reward and effort. Overall, the monkey preferred the lower-effort option (right panel). Sigmoidal fits provide intersection values for y=0.5 and x=0.5 to estimate reward- and effort-sensitivity. (F) Object-Sequence Learning. Learning the temporal order of objects is reflected in reduced errors, indexing successful error monitoring (left panel). The monkey choses incorrectly the order-irrelevant distractor at the temporal position at which it is similar to the correct target object for that slot, which was slot two for the initial sequence (blue line) and slot four 4 at a later encounter (red line). This indexes successful distractor interference control and serial item order learning.

Figure 7. Monitoring gaze and touch during task performance.

(A) Example horizontal and vertical traces of gaze (blue) and touch (red) behavior during task performance. Saccade and fixation onsets (solid and dashed vertical bars) are classified with a robust median thresholding algorithm (Voloh et al., 2019). (B) An example path of the Maze Learning task. (C) Example of a density heat map of gaze coordinates measured while recalling the trajectory shown in (B). The high gaze precision is enabled by calibrating gaze with an 9-point calibration routine built into -M-USE.